Chinese education companies hoping to profit from artificial intelligence are finding that investments in the technology can be as volatile as the markets they are betting on.

In June, Fenbi founder Zhang Xiaolong spoke about his gains from technology stocks during a lecture at Renmin University of China. He also told students they had few prospects beyond finding employment in the state sector. The remarks drew criticism, prompting an apology and his subsequent resignation as CEO.

Fenbi had made substantial investments. In the first half of 2026, its equity holdings rose from almost nothing to RMB 479 million (USD 71.6 million), including an exchange-traded fund providing twice the daily exposure to US semiconductor stocks.

When US semiconductor stocks plunged in early July, Fenbi sold its holdings but still incurred a loss of USD 8.3 million.

The episode foreshadowed a similar problem at another education company, TAL Education Group.

TAL’s second-quarter fiscal 2027 results, released October 29, showed revenue of about USD 995 million, up 15.5% year-on-year. Operating profit rose 145.4% to roughly USD 236 million, while gross margin reached 59.8%. Yet the company reported a net loss attributable to shareholders of USD 47.4 million, partly because of declines in the value of its investments.

The recovery also looks different from inside the company. Several employees told 36Kr that TAL cut jobs in April and May across departments including frontend development and its Xueersi Peiyou brand. Some units lost 30–50% of their staff, they said.

Investment gains complicate TAL’s recovery

In 2021, TAL shut down its academic tutoring business for children from kindergarten through ninth grade in response to China’s restrictions on after-school tutoring. It subsequently pursued three areas of growth: enrichment classes, vocational education, and learning devices.

By fiscal 2026, revenue had reached USD 3.009 billion, while net profit had increased more than fivefold year-on-year.

The recovery continued into fiscal 2027. In the first six months, revenue rose 22% to USD 1.753 billion, operating profit climbed 237.8% to about USD 373 million, and net profit attributable to shareholders increased 132.1% to about USD 361 million.

But the two quarters tell different stories.

In the first quarter, attributable net profit surged 1,204.3% to USD 408 million. Operating profit was USD 137.2 million, while other income, which included investment gains, reached about USD 405.2 million. That figure alone was nearly three times operating profit.

In the second quarter, other income swung to an expense of USD 189.7 million, pushing TAL into a net loss. Management attributed the change mainly to declines in the fair value of certain investments, warning that their valuations could continue to fluctuate with market conditions.

The company also recorded USD 135.9 million in income tax expense, with the increase largely attributable to changes in deferred tax liabilities related to withholding tax.

TAL invested in Z.ai, formerly Zhipu AI, in October 2023 and increased its investment the following year through affiliate Xinxin Xiangrong.

Z.ai listed in Hong Kong on January 8 this year, with an initial market capitalization of about HKD 51.1 billion (USD 6.5 billion). Its shares subsequently surged, reaching an intraday high of HKD 2,980 (USD 380) on June 22 and briefly lifting its market capitalization above HKD 1 trillion (USD 127.5 billion).

TAL’s stake in Z.ai through Xinxin Xiangrong amounted to just 0.79%, but the increase in the AI company’s valuation represented substantial unrealized gains.

Its investment portfolio extends beyond Z.ai. According to corporate information platform Tianyancha, TAL has also made strategic investments in BrainCo, Baichuan AI, StepFun, and DeepLang AI. Over the past five years, the former tutoring giant has built exposure to several AI companies.

TAL’s earnings release does not identify which investments accounted for the second-quarter valuation losses.

Meanwhile, its underlying education business remains profitable.

Enrichment classes have become a major source of growth. TAL offers mathematics-related courses under the label of “scientific thinking” and Chinese-language instruction as “humanities and creative writing.” Many parents view these classes as a way to supplement their children’s academic learning without enrolling them in courses explicitly categorized as academic tutoring.

At the end of the first fiscal quarter, TAL operated more than 600 offline learning centers across 44 cities in China and overseas markets. In the second quarter, revenue from offline learning services continued to grow at a double-digit rate year-on-year, while reenrollment in Peiyou enrichment courses remained above 80%.

Learning tablets face a shrinking market

TAL continues to invest heavily in learning devices, another pillar of its post-crackdown business.

Enrichment classes require more teachers as enrollment grows, limiting how far costs can fall as the business expands. Vocational education, meanwhile, is already served by established competitors such as Fenbi and Huatu Education.

Hardware offers a different approach. By incorporating its teaching materials into devices sold directly to households, TAL can generate revenue without expanding its teaching workforce at the same rate.

Early experiments with smart desks and lamps were constrained by limited repeat purchases. Learning tablets subsequently emerged as a more substantial business.

TAL now has more than two million learning devices active each week, representing a weekly activity rate of about 80%. Active devices are used for nearly an hour a day on average.

But growth in the broader market is slowing. According to research firm Runto, learning tablet sales in China fell 30.8% by volume and 17.1% by value across all channels in the third quarter. The firm expects full-year unit sales to decline 21.3% in 2026.

During TAL’s earnings call, president and CFO Alex Peng said intensifying competition, shifting consumer preferences, and rising costs for chips and other components had made learning devices a drag on overall revenue growth. The company expects its revenue growth to moderate.

Manufacturers are responding partly by pushing further into premium products.

In the first half of the year, flagship learning devices from Xueersi and iFlytek launched at prices above RMB 8,000 (USD 1,200), with some exceeding RMB 10,000 (USD 1,490). Whether parents will continue paying these prices is becoming an important question.

A Xinhua report published in 2025 estimated that a mainstream 14-inch learning tablet contained RMB 1,000–1,200 (USD 149–179) worth of hardware, with another RMB 200–400 (USD 29.9–59.8) per unit allocated to textbooks, question banks, and courses.

With limited scope to differentiate the hardware itself, manufacturers are turning to AI to support their premium pricing.

iFlytek launched its Spark model in May 2023. NetEase Youdao followed with Ziyue, Xueersi with MathGPT, and Yuanfudao with Kanyun. These products are marketed as AI-powered tutors that can provide continuous assistance and improve as the underlying models advance.

But general-purpose AI models are making that proposition harder to defend.

Youdao released its LobsterAI agent as open source in February. Tencent subsequently introduced its LearnBuddy education agent, while Baidu launched a similar product powered by Ernie.

Parents seeking help with a mathematics problem can increasingly use free or low-cost AI tools to obtain answers and step-by-step explanations. As comparable assistance becomes more accessible, manufacturers face growing pressure to justify the premium attached to dedicated learning devices.

Cost discipline brings workforce changes

Earlier in fiscal 2027, TAL made disciplined execution an operating priority. It reiterated that emphasis in its second-quarter discussion of learning center expansion, device costs, and resource allocation.

Selling and marketing expenses fell 4.8% year-on-year in the first quarter and 15.3% in the second. TAL was generating more revenue while spending less on those activities.

Employees described another side of the company’s cost controls.

In August, users of Xueersi learning tablets experienced problems logging into the user center. TAL attributed the disruption to backend maintenance and upgrades. One employee told 36Kr that reductions in the maintenance team may have contributed to the problem.

“A lot of problems look accidental, but they are actually inevitable,” the employee said.

Several employees said TAL had tightened performance reviews since 2022.

In 2024, it introduced what employees described as the “991 policy,” under which 90% of new hires would come through campus recruitment and 10% through experienced hiring, alongside a requirement to remove the lowest-rated 10% of employees through attrition.

“The compensation is N+1. People who refuse it receive a unilateral termination notice a month or two later,” one employee said.

N+1 generally refers to severance equivalent to one month’s salary for each year of service, plus an additional month’s pay.

According to one employee, the job cuts in April and May were more extensive than in previous years. Permanent employees could also be required to change roles.

Some teachers in the offline Peiyou business were asked to transfer to online Ledu courses.

“That effectively turns a full-time job into a part-time one. It’s a pay cut in another form.”

Employees could alternatively accept severance and return as outsourced workers, although few agreed to that arrangement, the employee said.

TAL’s published employment figures provide a broader picture. Its full-time headcount increased in each of the past three fiscal years, from about 15,000 in fiscal 2024 to 23,000 in fiscal 2025 and 26,100 in fiscal 2026. This indicates that the company continued expanding its overall workforce even as employees reported substantial cuts in certain departments.

Education companies search for new growth

China’s education companies face a common question in 2026: what can sustain investor interest after the tutoring crackdown?

Their responses have differed.

New Oriental has placed greater emphasis on returning earnings to shareholders. In July 2025, its board approved a three-year shareholder return plan that, beginning in fiscal 2026, commits at least 50% of the preceding year’s attributable net profit to dividends or share repurchases.

TAL has directed more capital toward AI investments, although it also announced its first cash dividend since listing alongside its latest quarterly results.

Gaotu has struggled to establish another source of growth. At the end of 2022, it launched Gaotu Jiapi, following East Buy into livestreaming commerce, but the initiative failed to develop into a substantial business. Gaotu has yet to launch its own foundation model or disclose significant progress in learning hardware.

Its first-quarter revenue rose 13.2% year-on-year to RMB 1.69 billion (USD 252.5 million), but net profit fell 72.2% to RMB 34.51 million (USD 5.2 million). In the second quarter, revenue increased 20.2% to RMB 1.67 billion (USD 249.6 million), while the company returned to a net loss of RMB 136 million (USD 20.3 million).

Smaller education companies are also experimenting with businesses far removed from their original operations.

One industry participant told 36Kr that their employer launched an AI-powered virtual try-on product in 2024.

“The boss would almost time experiments with new businesses to employees’ probation periods. Whatever was popular became the benchmark. The initial investment could run to tens of millions of RMB.”

Yuanfudao’s parent company has diversified into down jackets, the Grid Coffee chain, and postpartum care through Moli. Zuoyebang launched a romance game targeting women. Employees said the Baidu-backed company had also explored AI pets, food delivery, and medical aesthetics.

Such diversification has produced mixed results.

In July, IT training provider Itcast introduced courses in embodied intelligence, which involves AI systems interacting with the physical world, alongside six teaching robots designed for different levels of instruction. Its shares subsequently reached their daily trading limit for gains in eight consecutive sessions.

CEO Li Huoming wrote on WeChat that the IT training industry had “almost disappeared” over the preceding two years, describing the period as the darkest since the company’s founding. Itcast found a new direction, he added, after committing its resources to AI training in 2024.

Yet even as education companies seek growth from AI, policymakers are taking different approaches to its use in schools.

New York City has introduced a one-year moratorium on student-facing generative AI for the 2026–2027 school year, covering nearly 600,000 public school students from early childhood education through eighth grade.

Teachers may use approved AI tools for lesson planning and administrative work, but not for grading students. The policy includes exemptions for certain assistive technologies and permits limited, supervised AI use in high schools.

China, meanwhile, is pursuing broader adoption. Its national policy aims to establish widespread integration of AI into education by 2030, while schools are expected to use approved generative AI tools.

KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Xie Yunzi and Zhang Fan for 36Kr.

Note: HKD, RMB figures are converted to USD at rates of HKD 7.84 = USD 1 and RMB 6.69 = USD 1 based on estimates as of November 20, 2026, unless otherwise stated. USD conversions are approximate and, where appropriate, rounded for ease of reference. They may not fully match prevailing exchange rates.